Nitrogen (N) deposition and altered precipitation co-occur under global change, yet their interaction effects on soil organic carbon (SOC) and underlying microbial mechanisms remain unclear, especially in alpine grasslands. We conducted a coordinated N × precipitation manipulation experiment at two contrasting Tibetan Plateau sites: an arid alpine steppe and a humid alpine meadow. N addition increased SOC only under elevated precipitation, with no effect under ambient or reduced precipitation. The microbial pathways driving SOC accumulation differed by site: in the water-limited steppe, SOC gains were mainly associated with suppressed microbial carbon decomposition, whereas in the humid meadow, enhanced microbial carbon use efficiency (CUE) dominated, with smaller effects on decomposition. Ecoenzymatic stoichiometry showed that N addition reduced microbial carbon and N limitation, helping explain site-specific SOC responses. Our results indicate that precipitation not only controls the magnitude but also the mechanisms of N effects on SOC, highlighting the need to incorporate site-specific hydrological context and microbial carbon processing strategies into Earth system models.

Fertilization plays an important role in soil nutrient loss from sloping croplands. However, the effect of fertilization on Molybdenum (Mo) loss remains unknown. The aims of this study were to explore the effects of different fertilizers of purple soil on the characteristics of soil molybdenum loss in surface, subsurface runoff and sediments. Five fertilizers treatments (3 replicates) were designed as following: no fertilizer (CK); conventional nitrogen, phosphorus, and potassium fertilizer (NPK); organic fertilizers with livestock manure (OM); nitrogen, phosphorus, and potassium fertilizer plus organic fertilizers with livestock manure (OMNPK); and straw turnover plus nitrogen, phosphorus, and potassium fertilizer (RSDNPK). The changes of runoff-related Molybdenum loss from June to September 2025 were studied. Results showed that fertilization significantly reduced surface runoff and sediment yield compared with CK (p < 0.05). The RSDNPK treatment exhibited the lowest surface runoff, while OM and OMNPK treatments most effectively decreased sediment loss. Dissolved Mo (DMo) was the predominant form of Mo loss across all treatments (50 similar to 70% of total loss), significantly higher than particulate Mo (PMo, 25 similar to 40%) and Mo of soil sediments (SEMo, 6.5 similar to 12.9%). Notably, the OM treatment uniquely shifted Mo loss toward subsurface flow (47.2% of total), whereas other treatments were dominated by surface runoff. Total Mo loss amount varied significantly among treatments (p < 0.05): CK (795 mu g/m(2)) > OM (685 mu g/m(2)) > NPK (596 mu g/m(2)) > OMNPK (533 mu g/m(2)) > RSDNPK (373 mu g/m(2)). The RSDNPK treatment achieved the optimal performance, reducing total Mo loss by 53.1% compared with CK. Structural equation modeling revealed that soil organic matter indirectly controlled Mo loss by modifying soil physical properties and hydrological processes. The findings demonstrate that RSDNPK represents the most effective strategy for minimizing Mo loss in purple soil sloping croplands, outperforming sole organic manure application. This study highlights the importance of organic amendment and management in Mo loss control and provides a scientific basis for sustainable nutrient management in erosion-prone agricultural systems.

Understanding the spatial variability and environmental drivers of soil organic carbon (SOC) is critical for improving carbon management in fragile karst landscapes. This study collected 110 topsoil samples across county Yangshan, southern China, and applied an interpretable machine learning framework combining Random Forest (RF) and SHapley Additive exPlanations (SHAP) to explore the spatial heterogeneity and key environmental controls of SOC. The measured contents ranged from 3.33 to 44.20 g/kg, with a coefficient of variation of 43.5%, indicating moderate variability of SOC in the study area. The RF-based spatial predictions revealed that higher SOC levels were mainly concentrated in the northern and southern subregions associated with clastic rocks, while lower SOC values clustered in central areas dominated by carbonate bedrocks. SHAP analysis indicated that soil physicochemical properties contributed over 53% to SOC, with total nitrogen and cation exchange capacity exerting the strongest influences, particularly in karst zones. Hydrological, vegetation, and terrain-related factors showed moderate importance, especially in high-elevation areas with natural vegetation and complex topography that promoted SOC accumulation. In contrast, climatic variables had relatively weak impacts, with their influences clustered in lowlands dominated by anthropogenic land uses. These findings revealed spatially heterogeneous controls on SOC between karst and non-karst landscapes, emphasizing the dominant role of soil properties under shallow, erosion-prone conditions and highlighting the role of topography and vegetation in enhancing SOC stocks in mountainous areas. The integrated use of interpretable machine learning approaches improves the understanding of localized SOC dynamics and provides a valuable reference for precision carbon management and ecological restoration in environmentally sensitive regions elsewhere.

Molybdenum (Mo) is an important trace nutrient element in the soil and plays a significant role in maintaining plant growth. However, there are scarce studies on soil Mo content change and its driving factors based on historical soil samples. This paper studied the characteristics of Mo content in three different parent rock types (PRTs) and different eras. The findings indicated that the available Mo (AMo) and total Mo (TMo) in the purple soil were 0.087–0.131 mg/kg and 0.488–0.903 mg/kg, respectively, which were considered deficient. The TMo of J3p was higher than those of J2s and K2j, but the AMo was slightly lower than those of K2j and J2s. Compared with the old samples, the AMo of K2j, J2s and J3p has increased by 35.58%, 120% and 30.86%, respectively, and their TMo has increased by 29.37%, 25.21% and 11.97%, respectively. Our studies showed that PRTs directly impacted AMo, and indirectly influenced TMo and AMo through soil pH and organic matter. Organic matter and pH positively affected TMo, while pH negatively affected AMo. Overall, soil molybdenum content in the study area was generally insufficient, and local governments should comprehensively consider the molybdenum content and its main constraints for scientific fertilisation.

Accurate estimation of soil organic matter (SOM) content and its spatial distribution is crucial for sustainable agriculture and ecological management, Traditional SOM content measurement methods are insufficient, especially for high density vegetation areas in tropical and subtropical regions. The widespread use of unmanned aerial vehicles (UAVs) and numerous studies predicting soil information based on vegetation remote sensing data provide a solution to this problem. To evaluate the proposed approach. a 75-hectare agricultural field in Shaoguan City, Guangdong Province, was selected as the study arca. UAV hyperspectral imagery of vegetation at crop maturity was first acquired, and 103 soil samples were collected and transported to the laboratory for hyperspectral measurement and SOM content analysis, Subsequently, the continuous wavelet. transform (CWT) was applied to extract features from both the UAV vegetation and laboratory soil hyperspectral data. Finally, SOM content estimation and mapping were performed using random forest algorithms on the hyperspectral data before and after feature extraction, with the results compared to those obtained using Ordinary Kriging-based mapping. The results indicate that, (1) There is a significant correlation between UAV vegetation hyperspectral data and SOM content, although the accuracy ofSOM content estimation using UAV vegetation hyperspectral data was slightly lower than that using soil hyperspectral data; (2) After CWT, the accuracy of SOMcontent estimation using UAV vegetation hyperspectral data was superior to that of soil hyperspectral data, though still slightly lower than that of soil hyperspectral data after CWT; (3) The mapping accuracy of SOM content inversion using UAV vegetation hyperspectral data was better than that of the traditional Ordinary Kriging method, and was highly refined, Considering the significant advantages of UAV vegetation hyperspectral data in terms of cost and efficiency, this study suggests that the method of SOM contentestimation and mapping using UAV vegetation hyperspectral data is promising for providing abundant and detailed soil information for smart agriculture and other fields.

Similar to many other parts of the world, it is a necessity to reveal the spatial-temporal soil organic matter (SOM) dynamics of Mollisols in the northern Songnen Plain, a state key agricultural region in China. Although digital soil mapping (DSM) with temporal environmental covariates could fulfill this purpose, its accuracy still needs to be improved. The present study aimed to evaluate whether a spectral-temporal feature set derived from percentile transformations of time-series remote sensing images could be helpful for improving the accuracy, because the feature set is advantageous in providing wall-to-wall and stable information and having large dimensionality. The evaluation was conducted in the case of the Mollisols region, where a total of 334 soil samples were collected during 2009-2011 and 2014-2018 and measured for SOM contents. As environmental covariates, a spectral-temporal feature set consisting of a series of percentiles (i.e., 10 %, 25 %, 50 %, 75 %, and 90 %) of spectral bands and indices and corresponding means were derived from the MODIS/Terra images for every five years between 2009 and 2018, while the terrain and climate factors were also obtained. With these data, classification and regression tree (CART) and random forest (RF) were both employed to establish spatiotemporal models for predicting SOM content at five-year intervals from 2009 to 2018. Results showed that replacing the commonly used means and medians of spectral bands and indices in the two machine learning models with the spectral-temporal feature set improved the prediction accuracy, with an increase of the mean concordance correlation coefficient (CCC) by 1.94 %similar to 8.09 %. Further, the optimal RF model with the spectral-temporal feature set was used to generate SOM content maps for every five years between 2009 and 2018, which were validated based on the samples of 2009-2011, showing a CCC of 0.66. The resulted maps showed that the mean SOM content decreased from 2009 to 2018 by 0.04 %. An importance analysis showed that lots of the spectral-temporal features were the most important variables in the RF model, following the first important one (i.e., mean annual precipitation). As there were many spectral bands and indices, their sum importance was far larger than all the other kinds of environmental covariates. It is concluded that the spectral-temporal feature set is promising for deriving the spatial-temporal dynamics of soil in the future.

Accurate and high-resolution spatial soil information is crucial for efficient and sustainable land use, management, and conservation. Since the establishment of digital soil mapping (DSM) and the GlobalSoilMap working group, significant advances have been made in terms of the availability and quality of spatial soil information globally. However, accurately predicting soil variation over large and complex areas with limited samples remains a challenge, especially for China, which has diverse soil landscapes. To address this challenge, we utilised 11 209 representative multi-source legacy soil profiles (including the Second National Soil Survey of China, the World Soil Information Service, the First National Soil Survey of China, and regional databases) and high-resolution soil-forming environment characterisation. Using advanced ensemble machine learning and a high-performance parallel-computing strategy, we developed comprehensive maps of 23 soil physical and chemical properties at six standard depth layers from 0 to 2 m in China at a 90 m spatial resolution (China dataset of soil properties for land surface modelling version 2, CSDLv2). Data-splitting and independent-sample validation strategies were employed to evaluate the accuracy of the predicted maps' quality. The results showed that the predicted maps were significantly more accurate and detailed compared to traditional soil type linkage methods (i.e. CSDLv1, the first version of the dataset), SoilGrids 2.0, and HWSD 2.0 products, effectively representing the spatial variation of soil properties across China. The prediction accuracy of soil properties at all depth intervals ranged from good to moderate, with median model efficiency coefficients for most soil properties ranging from 0.29 to 0.70 during data-splitting validation and from 0.25 to 0.84 during independent-sample validation. The wide range between the 5 % lower and 95 % upper prediction limits may indicate substantial room for improvement in current predictions. The relative importance of environmental covariates in predictions varied with soil property and depth, indicating the complexity of interactions among multiple factors in the soil formation processes. As the soil profiles used in this study mainly originate from the Second National Soil Survey of China, conducted during the 1970s and 1980s, they could provide new perspectives on soil changes, together with existing maps based on soil profiles from the 2010s. The findings of this study make important contributions to the GlobalSoilMap project and can also be used for regional Earth system modelling and land surface modelling to better represent the role of soil in hydrological and biogeochemical cycles in China. This dataset is freely available at https://www.scidb.cn/s/ZZJzAz (last access: 17 November 2024​​​​​​​) or https://doi.org/10.11888/Terre.tpdc.301235 (Shi and Shangguan, 2024).

Soil organic matter (SOM) determines soil fertility and functions, playing a key role in agriculture, the environment and climate change. During the past century, the SOM of the world, e.g., the black soil (Mollisol) in croplands of Northeast China, experienced extensive changes, making SOM monitoring crucial. Recently, digital soil mapping (DSM) with time-series remote sensing images has become a mainstream method for SOM monitoring, but there is room for its accuracy to be improved. To fulfill this purpose, we propose utilizing crop residue indices (CRIs) derived from remote sensing images within the method, as crop residues are a main source of the SOM. In this study, performances of five commonly used CRIs, e.g., normalized difference tillage index (NDTI), on SOM monitoring was evaluated based on a series of topsoil samples collected from 2014 to 2018 in croplands of the center black soil region in Northeast China. The performances and those of cumulative CRIs computed over some years were compared to those of basic climate and terrain attributes, spectral bands, an empirical index, and commonly used vegetation indices (VIs, e.g., normalized difference vegetation index (NDVI)). Results showed that temporal CRIs had a stronger correlation with SOM content (0.52-0.73) than did the others (0.04-0.69). Integrating CRIs with basic soil covariates increased prediction accuracy by 7.27 % in Lin's concordance correlation coefficient (CCC). Further, the CRIs and VIs accumulated over 3 and 4 years, respectively, had a much stronger correlation with SOM (0.65-0.73 and 0.67-0.69, respectively) and led to better accuracies with an average increase of 2.62% in CCC compared to indices of the current sampling year. While annual SOM maps predicted with and without the optimal cumulative CRI showed similar spatial patterns, they were statistically significantly different. It is recommended to utilize the cumulative NDTI for monitoring SOM.

Simultaneous estimation of multiple soil properties from vis-NIR hyperspectra presents a cost-effective and timeefficient approach. Previous studies have utilized multi-task convolutional neural network (multi-CNN) with share-bottom structures based on the hard parameter sharing. However, multi-CNN often ignores the differential characteristics of correlations between soil properties, limiting the accuracy of soil property estimation. The multi-gate mixture-of-experts network (MMoE) offers a solution by extracting both common features across all soil properties and unique features specific to each soil property, which probably could provide better estimation outcomes than the conventional shared-bottom multi-CNN. In the present study, a MMoE was built based on a total of 17,272 mineral soil samples from the Land Use/Cover Area Frame Survey (LUCAS) topsoil database that includes vis-NIR spectra with ten physicochemical properties, i.e., clay, silt, sand, pH (in water), organic content (OC), calcium carbonate (CaCO3), nitrogen (N), phosphorous (P), potassium (K), and cation exchange capacity (CEC). To evaluate the performance of MMoE, a series of other models were also built, i.e., partial least square regression (PLSR), single-task convolutional neural network (single-CNN), multi-task convolutional neural network (multi-CNN) and multi-task long short-term memory (multi-LSTM). Furthermore, performance of feature-spectrum selected by competitive adaptive reweighted sampling (CARS) on the accuracy of the MMoE was also explored, as well as a data augmentation method of stacking raw spectra with five preprocessed spectra data. The results demonstrated that MMoE had higher accuracy than PLSR, single-CNN, and multi-LSTM models, with RMSE reduction of 5 %-48 %, R2 improvement of 1 %-119 %, and CCC improvement of 0 %-74 %. Compared with multi-CNN, MMoE showed better accuracy for all properties except pH, with RMSE reduction of 3 %-8 %, R2 improvement of 1 %-12 %, and CCC improvement of 0 %-5 %. However, the feature-spectrum selected by CARS did not improve the accuracy of MMoE compared to full-band spectrum, whereas the data augmentation method was effective in improving the estimation accuracy of MMoE compared to raw spectra, with RMSE reduction of 14 %-28 %, R2 improvement of 3 %-88 %, and CCC improvement of 1 %-63 %. Consequently, this study proves that MMoE based on data augmentation is an efficient and accurate method for the simultaneous estimation of multiple soil properties from vis-NIR spectra.

Abstract. Accurate and high-resolution spatial soil information is crucial for efficient and sustainable land use, management, and conservation. Since the establishment of digital soil mapping (DSM) and the GlobalSoilMap working group, significant advances have been made in spatial soil information globally. However, accurately predicting soil variation over large and complex areas with limited samples remains a challenge, especially for China, which has diverse soil landscapes. To address this challenge, we utilized 11,209 representative multi-source legacy soil profiles (including the Second National Soil Survey of China, World Soil Information Service, First National Soil Survey of China, and regional databases) and high-resolution soil-forming environment characterization. Using advanced Quantile Regression Forest algorithms and a high-performance parallel computing strategy, we developed comprehensive maps of 23 soil physical, chemical and fertility properties at six standard depth layers from 0 to 2 meters in China with a 90 m spatial resolution (China dataset of soil properties for land surface modeling version 2, CSDLv2). Data-splitting and independent samples validation strategies were employed to evaluate the accuracy of the predicted maps quality. The results showed that the predicted maps were significantly more accurate and detailed compared to traditional soil type linkage methods (i.e., CSDLv1, the first version of the dataset), SoilGrids 2.0, and HWSD 2.0 products, effectively representing the spatial variation of soil properties across China. The prediction accuracy of most soil properties at the 0–5 cm depth interval ranged from good to moderate, with Model Efficiency Coefficients for most soil properties ranging from 0.75 to 0.32 during data-splitting validation and from 0.88 to 0.25 during independent sample validation. The wide range between the 5 % lower and 95 % upper prediction limits may indicate substantial room for improvement in current predictions. The relative importance of environmental covariates in predictions varied with soil properties and depth, indicating the complexity of interactions among multiple factors in the soil formation processes. As the soil profiles used in this study mainly originate from the Second National Soil Survey of China during 1970s and 1980s, they could provide new perspectives of soil changes together with existing maps based on 2010s soil profiles. The findings make important contributions to the GlobalSoilMap project and can also be used for regional Earth system modeling and land surface modeling to better represent the role of soil in hydrological and biogeochemical cycles in China. This dataset is freely available and can be accessed at https://doi.org/10.11888/Terre.tpdc.301235 (Shi et al, 2024).

Sampling density and depth play crucial roles in three-dimensional (3D) soil modeling and prediction, particularly in digital soil mapping (DSM). However, previous studies have yielded inconsistent and even contradictory results regarding impacts of sampling density and depth on the accuracy of 3D DSM. Hence, this study aimed to evaluate the impacts of sampling depth, vertical sampling density and lateral profile density on 3D soil mapping accuracy based on a case study of soil organic carbon (SOC). To achieve this, a comprehensive analysis was conducted based on 511 soil samples collected from 111 profiles in a local hilly area spanning 5.52 km2 in China. A 3D regression geostatistical approach was employed for the analysis. The results revealed that samples taken from different depth intervals exhibited varying degrees of importance in relation to prediction accuracy. Notably, a reduced number of surface soil samples (0–0.3 m) led to significant fluctuations in the accuracy of predictions for the entire soil profile. Furthermore, a lower number of subsurface soil samples (0.3–0.6 m) also diminished the overall prediction accuracy, while the influence of deeper soil samples (0.6–1.2 m) on the overall accuracy was relatively less pronounced. Reducing the number of profiles for calibration led to significantly worse and more variable prediction accuracy, compared to reducing vertical samples. For 3D mapping, this study recommends prioritizing the collection of additional lateral profiles, specifically focusing on surface and subsurface soils, especially for properties that display vertical variation similar to SOC.

AbstractThe countries of Central Asia are collectively known as Uzbekistan, Kyrgyzstan, Turkmenistan, Tajikistan and Kazakhstan. Central Asian countries have experienced significant warming in the last century as a result of global changes and human activities. Specifically, the five Central Asian countries’ populations and economies have increased, with Turkmenistan showing the fastest growth rates in GDP and per capita GDP. Farmland change, forestry activities, and grazing are examples of land use/land cover change and land management in Central Asia. Land degradation was primarily caused by rangeland degradation, desertification, deforestation, and farmland abandonment. The raised temperature, accelerated melting of glaciers, and deteriorated water resource stability resulted in an increase in the frequency and severity of floods, droughts, and other disasters. The increase of precipitation cannot compensate for the aggravation of water shortage caused by temperature rise in Central Asia. The ecosystem net primary productivity was decreasing over the past years, and the organic carbon pool in the drylands of Central Asia was seriously threatened by climate change. Grassland contributed the most to the increase of ecosystem service values in recent years. Most ecosystem functions decreased between 1995 and 2015, while they are expected to increase in the future (except for water regulation and cultural service/tourism). Global climate change does pose a clear threat to the ecological diversity of Central Asia.

Accurate and effective monitoring of potentially toxic elements (PTEs) in soil across vast regions is crucial for environmental modeling and public health. While remote sensing (RS) technology provides a promising approach by detecting soil spectrum, dense and persistent vegetation cover in subtropical agricultural areas hinders acquisition of bare soil signals, limiting soil PTEs monitoring. To address this challenge, the present study proposed an innovative method for monitoring soil arsenic (As) content by using vegetation characteristics retrieved from RS data as proxy variables, given soil-vegetation interactions. The method was evaluated in a densely vegetated cropland of southern China, where 104 surface soil samples were collected. Vegetation information was extracted both individually and synergistically using time-series Sentinel-2 multispectral and Sentinel-1 synthetic aperture radar (SAR) images throughout the entire growing season, and an unmanned aerial vehicle (UAV) hyperspectral image during the crop maturity. Multiple machine learning algorithms, including Random Forest, Support Vector Regression, CatBoost, and Stacking were applied to model the relationship between soil As and vegetation variables. The SHapley Additive exPlanation (SHAP) technique was introduced for identifying key variables and corresponding thresholds indicating significant accumulation of soil As. Results showed that time-series satellite-multispectral images outperformed other single RS data types in terms of prediction accuracy. Moreover, the synergy of optical and SAR images significantly improved model accuracy. Particularly, the combination of time-series satellite multispectral and SAR data using the stacking algorithm achieved the best results, with a coefficient of determination (R2) of 0.71 and a root mean square error (RMSE) of 20.22mg/kg. Key predictive variables included red-edge vegetation index (RENDVI3) on August 7 and May 26, and the blue band on October 26, with values below 0.018, 0.013 and 0.052, respectively, indicating the As accumulation in soil. In summary, the proposed method of using multiple RS data to retrieve vegetation characteristics for inferring soil PTEs in densely vegetated areas was convenient, cost-effective, and reliable, offering new insights and technical support for environmental monitoring.

[Objective] In order to investigate the influencing factors and variation rules of accuracy of various two-step models of 3D mapping methods. [Methods] Soil organic carbon in a forest with an area of about 5 km2 in a typical hilly region of South China was mapped. Spline functions, exponential functions and power functions were used as depth functions, ordinary kriging and random forest were used as horizontal mapping methods and two different mapping forms (called forms A and B) were used. The 3D prediction mapping of soil organic carbon was carried out, and the influence of different depth functions, horizontal mapping methods and mapping forms on the 3D mapping results of two-step model was explored. [Results] (1) The depth function largely determined the variation of mapping results in vertical and horizontal directions, which showed that the variation of the mapping results was significantly different among the three depth functions. Exponential function had the largest variation and power function had the weakest variation, while horizontal mapping methods (ie.ordinary kriging and random forest) had little influence on the vertical variation of mapping results. However, the spatial variation of surface layer was greatly affected by the horizontal mapping methods. (2) The accuracy of spline function was the best because the simulated depth curve was in the best agreement with the measured values. The consistency correlation coefficients (CCC) of the 3D mapping based on the spline function were 0.72 and 0.75, which were higher than the other functions in the same form of 3D mapping (CCC were between 0.64 and 0.74). For the horizontal mapping methods, the accuracy of the ordinary kriging was better than that of the random forest, the CCC of the former was between 0.67 and 0.75, the latter was between 0.64 and 0.72; (3) The two mapping forms of the two-step model had little influence on accuracy. Only in the case of prediction for bottom layer, form A (i.e., horizontally mapping simulated walues of a depth function was better than form B (i.e., simulating parameters of a depth function).(4) Among all 3D mapping methods, the form A, with ordinary kriging and spline functions, had the highest accuracy, generating coefficient of determination (R2) of 0.76, CCC of 0.75, and root mean square error (RMSE) of 3.50 g/kg. [Conclusion] In the two-step model of 3D soil mapping, firstly, the spline function should be considered as the depth function. Secondly, the horizontal mapping method should be considered according to the landscape conditions and sample size. Finally, the first mapping form of the two-step model should be adopted as far as possible.

The generalized linear geostatistical model (GLGM) is a formal approach of regression kriging that would be advantageous over commonly used modelling approaches for digital soil mapping (DSM). However, it has not been well explored in the literature, due to heavy computation. This study evaluates such formal approach for mapping soil organic matter at a regional scale (179,700 km2). We hypothesized that GLGM would be a better approach than other approaches as it can model nonlinear relationships and spatially consider the residuals. However, the accuracy would depend on sampling density. We compared GLGM with multiple linear regression (MLR), ordinary kriging (OK), regression kriging (RK), random forest (RF), generalized linear mixture model (GLMM), and generalized additive model (GAM). The effect of sampling density on the performance was also investigated by fitting the models based on resampling the samples with a series of sizes ranging from 100 to 1200. Results showed that GLGM generally improved the accuracy of DSM, compared with MLR, OK, RF, GLMM, and GAM, especially for large sample sizes, although the improvement was not significant. In a few cases, GLGM outperformed RK. The GLGM modelling and its prediction were largely influenced by sampling densities. Given small sample sizes, GLGM was unstable and the parameters were highly variable depending on the modelling realizations. Other influencing factors include linear and smooth correlations between soil and environmental covariates, spatial autocorrelation of the residuals, compatibility of spatial scales of soil samples and environmental covariates, and the scale of soil variation. For these factors, researchers with a more dense soil sampling are needed in future to explore the benefits of GLGM as a hybrid model for spatial prediction of soil.

Accurately monitoring soil organic matter (SOM) content is crucial for food and soil security. Current methods of monitoring are expensive and existing sparse data cannot provide detailed spatial information about SOM content changes in an area. This study proposes using a spatiotemporal model with time-series synthetic Landsat images to monitor SOM content dynamics at the regional scale. The approach was implemented in Google Earth Engine (GEE) platform and tested in Jiangsu province, China, using the soil survey data from 2006 to 2007 and synthetic Landsat images from 1986 to 2007. The model generated SOM maps every three years between 1986 and 2007 and was evaluated using another soil survey from 2000 in southern Jiangsu. The model associated with 20 covariates derived from the synthetic Landsat image explained 70% of the variation in SOM content with root mean squared error (RMSE) and Lin's concordance correlation coefficient (CCC) of 5.17 g/kg and 0.57, respectively. The results showed that the model could reveal regional spatial and temporal differences in SOM content distribution. The SOM content in Jiangsu increased in the north, while decreasing in the central and southern areas. Temporally, the mean SOM contents increased from 1986 to 1992 by 0.17 g/kg, decreased in 1995, and increased again from 1998 to 2000 before decreasing from 2004 to 2007 by 0.14 g/kg. The validation based on the soil data in 2000 showed that the approach generated an RMSE of 5.97 g/kg, accounting for 22.77% of the average SOM content of the data. The study concluded that this approach could be used for monitoring SOM content and other soil properties. This approach had a relatively better accuracy than a previous study using the Integrated Nested Laplace Approximation with the Stochastic Partial Differential Equation approach with the same soil data but 4 times more samples.

Increased soil organic carbon (OC) in China has been reported in the past two decades, suggesting the sequestration of atmospheric carbon dioxide into soil, mitigating climate change and improving soil health. On the other hand, soil pH decrease had also been reported nationwide. If the two are related, the strategy of increasing soil OC could negatively affect soil quality for food production and the environment. We investigate this thread based on large-scale soil survey data from two provinces with typical soil and cropping patterns in the east and south of China, Jiangsu (102,600 km2) and Guangdong (177,900 km2). The data include >5000 observations from soil surveys conducted over the past four decades, i.e., the 1980s, 2006-2007, and 2010-2011. Using spatiotemporal modelling, we show that across Jiangsu province, the topsoil OC on average has increased from 8.5 g kg-1 to 9.9 g kg-1 from 1980 to 2000 and a further increase to 12.6 g kg-1 in 2010. This increase was accompanied by a decrease in average pH from 7.63 to 6.90. In Guangdong, there was an overall increase in average topsoil OC content from 14.2 g kg-1, 16.5 g kg-1, and 20.2 g kg-1 with a decrease in average pH from 5.58, 4.90, and 4.98. Based on the spatiotemporal modelling results, the structural equation modelling analysis shows that OC and pH changes were significantly correlated and linked by increased soil N content. On croplands, soil N content was mainly attributed to N fertiliser application. The pH decrease was particularly significant in the east of China where the soils were neutral in pH. We recommend that more revolutionary means be taken to sequestrate atmospheric carbon into soil as the current OC increase due to increasing crop productivity via a high rate of nitrogen application may have a potential acidification effect.

It is attractive nowadays to estimate soil information based on in situ Visible and Near-infrared (Vis-NIR) spectroscopy. However, there exist a lot of errors, mainly due to modeling between soil properties and spectra. A hybrid of deep learning (DL) models seems an ideal approach for the modeling, e.g., convolutional neural network (CNN) and long short-term memory (LSTM). Nevertheless, it is not easy for a user, or not efficient for an automatic technique, such as Bayesian optimization (BO), to select hundreds of parameters of a hybrid of DL models. Recently, tree Parzen Estimator (TPE) based HyperBand within BO (BOHB) has been widely used in other fields for establishing a hybrid of DL models efficiently and automatically. Further, it is not quite clear if a DL model or a hybrid DL model performs well for different sample sizes and different numbers of spectral data. This study aimed to investigate these issues by evaluating performances of two hybrids of DL models (CNN-LSTM and LSTM-CNN) established using BOHB, based on soil spectra and organic matter contents of 670 soil samples collected from a forest in Nanning, southwest China. The performances were also compared with those of separate DL models and two commonly used methods, i.e., partial least square regression (PLSR) and random forest (RF), while the effect of sample sizes on the performances was also investigated. Besides, the effect of different numbers of spectra was also investigated, and the spectra were augmented using two methods, one by stacking scans of all measurement points of a sample and one by stacking preprocessed spectra using several methods. Results showed that, given less than 600 samples, accuracies of CNN-LSTM and LSTM-CNN were not higher than those of separate DL models, PLSR and RF. However, BOHB optimization largely improved the accuracies of the two hybrid models which were close to or higher than those of separate DL models, PLSR and RF. With the increase of sample sizes, accuracies of all models used in this study gradually increased and the DL models seemed to be more promising for large sample sizes. Furthermore, augmenting spectra with preprocessed data provided much more benefits than sample sizes, modeling methods, and optimization methods. Thus, it is promising to estimate soil using in situ Vis-NIR spectra based on a hybrid of DL models optimized using BOHB, particularly for large datasets or augmented spectra.

The accuracy of digital soil mapping (DSM) is related to many elements of a soil survey. However, most past studies commonly focus on one or two elements of DSM related to sample sizes and modeling methods, ignoring the effect of the soil-forming environment. For instance, different landform types can cause striking soil varia-tions. As a result, studies often present different or even contradictory relationships between sample sizes and mapping accuracy. This study evaluated the accuracy of DSM in predicting soil organic carbon content due to the interaction of sample sizes, modelling methods and two landform types (i.e., hill-mountains and plains), based on 1861 soil samples from an existing soil survey conducted in the Guangdong province of China (about 180 000 km(2)). The study evaluated six modeling methods, including multiple linear regression, ordinary kriging, regression kriging, quantile regression forest, geographical weighting regression and radial basis function. The sample sizes varied from 100 to 1300 with an interval of 100. Results show that generally mapping accuracy increased with the sample sizes. However, this relationship was often disrupted by randomness in sampling procedure. Further, the different modeling methods or different landform types could lead to further variation, while the general relationship did not change. Thus, this study demonstrates that variation of mapping accuracy was collectively affected by many elements of DSM: sample sizes along with randomness in sampling, modelling methods and landform types, and differed between different accuracy metrics. The collective effect is the reason for different or contradictory relationships of mapping accuracy reported in the literature, where only one or two elements were investigated. For the collective effect, the relationship of mapping accuracy with elements of DSM could be unique in an area.

Drought is a meteorological phenomenon that threatens ecosystems, agricultural production, and living conditions. Central Asia is highly vulnerable to drought due to its special geographic location, water resource shortages, and extreme weather conditions, and poor management of water resources and reliance on irrigated agriculture exacerbate the effects of drought. In this study, the latest version of the Global Land Data Assimilation System was employed to calculate the Standardized Precipitation Evapotranspiration Index at different time scales during the period from 1981 to 2020. The varimax Rotated Empirical Orthogonal Function was applied for subregional delineation of drought patterns in Central Asia, and various methods were employed for a comparative analysis of the spatiotemporal characteristics of drought in these Central Asian subregions. The results show that drought patterns vary considerably in the Central Asian subregions. Over the past 40 years, alternating wet and dry conditions occurred in Central Asia. North Kazakhstan experienced more drought events with lower severity. East and west differences appear after 2001, the west becoming drier and the east becoming wetter. Some regions near lakes, such as Balkhash, Issyk-Kul, and the Aral Sea, suffer from droughts of long duration and high severity. In the Tianshan region, droughts in the northern slopes occur more frequently, with shorter durations and higher intensity and peaks. Northwestern China and western Mongolia have extensive agricultural land and grasslands with highly fragile ecosystems that have become progressively drier since 2001.